1

Paper 5206-2020

The Plight of the Honeybee

Rachael Bishop, DaMarkus Green, Karin Kolb

Kennesaw State University

ABSTRACT

A decline in the honeybee population is raising concerns worldwide. Honeybees are an

important part of agricultural industries. Is it possible that the neonicotinoid pesticides used

to protect crops from damaging insects are also harming the insects necessary for

pollinating the same crops? This study explores the effects of these pesticides on honeybees

by comparing annual honey production yields and honeybee colony counts in the United

States from 1995 through 2015.

INTRODUCTION

It is becoming common knowledge that the bees we normally swat away have a greater

importance than once imagined. No longer seen as pests, they are emerging as unsung

heroes actively pollinating the majority of our food crops for as long as they have buzzed

the Earth. Out of the 20,000 known bee species in the world, 4,000 of them are native to

the United States. It is estimated that 80 percent of flowering plants and crops around the

world are pollinated by native bees; as well, 75 percent of the fruits, nuts, and vegetables in

the United States are pollinated by a variety of bees (Lubeck, 2019). Overall, an annual

increase of $14 billion in crop value can be attributed to honeybee pollination in the U.S.

(“Pesticide toxicity to bees”, 2019).

With the recent acknowledgment of the endangerment of bees and the decrease of their

yearly pollination, it is of utmost importance to monitor bee activity in an attempt to

increase their population and their pollination efforts in the United States and worldwide.

The usage of pesticides on our nation’s crops, in an attempt to increase crop output, has

potentially done more damage than good. Injuring and killing our nation’s natural

pollinators may be permanently damaging our agricultural output. Time is of the essence to

determine what is really causing our natural pollinators, our bees, to die off, resulting in an

overall reduction in pollination and in production of our nation’s food sources.

One place to begin, is investigating honeybees and one of their top executioners: pesticides.

Because honeybees leave evidence of their presence through the production of honey, there

is a measurable variable to analyze and draw conclusions about the viability of this species

of bees. Neonic pesticide usage began in the 1990’s; however, with honey production

declining years earlier, it is of interest to look into whether the use of these pesticides has

increased the ongoing decline of honeybee honey production in the United States.

BACKGROUND

Neonicotinoid (neonic) pesticides are insecticides derived from nicotine. In June of 2017,

two large-scale studies confirmed that in agricultural areas, neonic pesticides are hurting

the ability for bees to reproduce, as well as, killing them. It was argued that previous

studies used unrealistic quantities of pesticides to substantiate their hypothesis. The new

studies claim the damage is done over an extended period of time (Dengler, 2017). 31.6%

of honeybees that are exposed to pesticides fail to return to their hives, while the remaining

bring back contaminated pollen which in turn affects the other workers and the queen at the

hive (“Pesticide toxicity to bees”, 2019). Understanding the effects of neonic pesticides on

the health of honeybees is necessary to reduce the impact on food sources and lives of

bees.

2

DATA

The data set, Honeybees and Neonic Pesticides, used for the analysis is provided by the

National Agricultural Statistics Service (NASS). The U.S. Department of Agriculture (USDA)

uses the NASS as their primary data reporting source. The data set also includes results

from the U.S. Geological Survey (USGS) Pesticide National Synthesis Project. Spanning 20

years, 1995 to 2015, it provides statistics regarding honey production supply and demand in

most of the states in the U.S., as well as detailed information on Neonicotinoid (neonic)

pesticides. There is no data for Alaska, Connecticut, Delaware, Massachusetts, New

Hampshire and Rhode Island. This data set is available through www.kaggle.com and was

compiled by Kevin Zmith.

The set includes seventeen variables and 1132 observations. The variables of particular

interest in this study are State Name, Region, Number of Colonies, Total Production, which

is the number of honey-producing colonies multiplied by the honey yield per colony in a

given U.S. state per year, and the five neonic pesticides, Clothianidin, Imidacloprid,

Thiamethoxam, Acetamiprid, and Thiacloprid. The full list of variables and their descriptions

can be found in the data dictionary in Appendix A Table 1.

PROBLEMS ADDRESSED

The USDA provides a very informative website full of information pertaining to the

agriculture industry. There is a section dedicated to their Pesticide Data Program. The

addition of an interactive tool, such as a decision tree, could be a resource for concerned

growers to access when making a decision about pesticide usage. This predictive model

could help growers make informed decisions about the use of specific neonic pesticides and

the potential consequences to honeybees, pollination, and their bottom line, crop production

yield.

The National Farmers Union is an organization whose mission is to advocate for U.S.

farming families through education, cooperation and legislation. They provide important

agricultural updates pertaining to all farmers. Sharing the results of a comparison model on

their website provides valuable information for interested farmers to determine the effects

in their region from the use of specific pesticides. Using this model, future honey yield and

colony count could be estimated to better monitor and track the decline of honeybee

populations.

DATA CLEANING AND PREPROCESSING

Observations dated before 1994 are removed; although they included data about annual

honey yield, they did not include data about neonic pesticide use. The observations for 1994

are removed once the new target variable is created. The observations for the state of

Hawaii are removed as well for lack of data about neonic pesticide usage. The resulting 856

observation data set contains data about 43 out of the 48 contiguous United States.

Prior to developing a decision tree model, a new target variable, Effect, is created

comparing the total production per state to each preceding year to see if there is a decrease

in production caused by an increase in neonic pesticide usage. This comparison is made by

comparing two additional new variables, ProdDecr and NeonIncr.

ProdDecr is a binary variable assigned by comparing a year’s total honey production to the

previous year’s total honey production for each year in each state. The total honey

production in pounds for 1994 is the control observation for the 1995 comparison. If there is

a reduction in the total pounds produced then ProdDecr is assigned a value of one. If there

is an increase or no change in pounds produced then ProdDecr is assigned a value of zero.

NeonIncr is a binary variable assigned by comparing a year’s total neonic pesticide usage to

the previous year’s total neonic pesticide usage for each year in each state. Neonic pesticide

3

usage data is not recorded prior to 1995, so a quantity of 0kg is the control for the 1995

comparison. If there is an increase in total kg pesticide usage then NeonIncr is assigned a

value of one. If there is a reduction or no change in total kg pesticide usage then NeonIncr

is assigned a value of zero.

The target variable, Effect, is a binary variable assigned by comparing the values of

ProdDecr and NeonIncr. If both ProdDecr and NeonIncr have a value of one, then Effect is

assigned a value of one based on the assumption that total honey production is negatively

affected by an increase in pesticide usage. Any other combination of values for ProdDecr

and NeonIncr results in an assignment of the value zero for the variable Effect.

ANALYSIS

SAS® ENTERPRISE MINER™

The data set including the new target variable is imported into SAS® Enterprise Miner™. It

is partitioned into 80% training and 20% validation. Redundant variables such as state and

StateName are reduced to one selected input variable. All of the neonic pesticide variables

are positively skewed distributions in need of transformation. Three possible

transformations are explored: the ‘Best’ transformation as chosen by Enterprise Miner, the

Log transformation, and the Square Root transformation. Attempts to create a useful

decision tree model are unsuccessful. Among those attempted, a 5 Branch Chi-Squared Tree

is the only model that produces a tree. However, the variables THIAMETHOXAM and

IMIDACLOPRID are the only ones of importance. Figure 1 shows the resulting decision tree.

Figure 1. Five Branch Chi-Squared Tree

Unfortunately, our decision tree fails to produce results to address our first problem

regarding the USDA and the use of specific pesticides. We continue our study by conducting

further analysis to address our second problem. We endeavor to create a model to aid

concerned farmers who access the National Farmers Union website.

SAS®

SAS Proc Reg is applied to build the predictive models. The analysis includes all of the

individual neonic pesticide variables, as well as, numcol and totalprod, filtered by Region;

South, West, Midwest, and Northeast. These regions all have area-specific characteristics

that may be of interest for consideration in future research. After dividing the data set by

Region, SAS regression analyses are run to determine if any of the specific pesticides have a

4

significant impact on total honey production yield or honeybee colony count. The parameter

estimates enable the building of a model based on individual pesticides, facilitating the

calculation of estimated effect on total honey production and colony count. The output for

each of the regions is shown below in Output 1 through Output 4.

Output 1. South Region: Total Honey Yield Model and Total Colony Count Model

Output 2. West Region: Total Honey Yield Model and Total Colony Count Model

Output 3. Midwest Region: Total Honey Yield Model and Total Colony Count Model

Output 4. Northeast Region: Total Honey Yield Model and Total Colony Count Model

5

Time series plots are developed for each region based on the total yield of honey each year

to provide a visual depicting the possible effect of pesticide usage in a sampling of states in

each region. The plots are shown in Appendix B Figure 2 through Figure 5.

CONCLUSION

For our addressed problems, analyzing the use of neonic pesticides on U.S. crops starting in

1995, the expectation was that all of the pesticides have a negative effect on both total

honey production yield and total honeybee colony count. Contrary to this belief, the findings

suggest that the pesticides THIAMETHOXAM and THIACLOPRID have a positive effect on

both yield and colony count. The farming community may find the regression model useful

in determining which neonic pesticides are beneficial and which are harmful in their specific

region, however, more research is needed. Both our decision tree and regression models

identified that if the quantity of THIAMETHOXAM used is monitored, it may result in a

positive impact on honey production yield and honeybee colony counts, and subsequently

overall crop yield. All the other neonic pesticides were seen as harmful to both honey

production and colony count and therefore should be avoided, if possible.

For future research it would be interesting to look into weather patterns, soil acidity, and

pollen counts to determine if there is a significant difference between these regions and if

these factors have an impact on the results found through this study.

REFERENCES

Dengler, R. (2017, June 29). Neonicotinoid pesticides are slowly killing bees.

Retrieved from PBS News Hour: https://www.pbs.org/newshour/science/neonicotinoid-pesticides-slowly-

killing-bees Lubeck, M. (2019, June 20). Honey Bee Helpers: It Takes a Village to Conserve a

Colony. Retrieved from U.S. Geological Survey:

https://www.usgs.gov/news/honey-bee-helpers-it-takes-village-conserve-colony

Pesticide toxicity to bees. (2019, November 11). Retrieved November 16, 2019, from Wikipedia: https://en.wikipedia.org/w/index.php?title=Pesticide_toxicity_to_bees&oldid

=925625965. Zmith, K. (2018, May 16). Honeybees and Neonic Pesticides. Retrieved from

kaggle: https://www.kaggle.com/kevinzmith/honey-with-neonic-pesticide

CONTACT INFORMATION

Your comments and questions are valued and encouraged. Contact authors at:

Rachael Bishop DaMarkus Green

rbisho16@students.kennesaw.edu dgreen62@students.kennesaw.edu

Karin Kolb Dr. Sherry Ni – Faculty Advisor

kkolb1@students.kennesaw.edu sni@kennesaw.edu

6

APPENDIX A

Variable Description

state Abbreviated state name

numcol Number of honeybee colonies in the state

yieldpercol Honey yield per colony. Unit is pounds

totalprod Total production (numcol x yieldpercol). Unit is pounds

stocks Refers to stocks held by producers. Unit is pounds

priceperlb Refers to average price per pound based on expanded sales. Unit is dollars

prodvalue Value of production (totalprod x priceperlb). Unit is dollars

year Recording year

StateName Full state name

Region U.S.-Census-classification of states

FIPS Federal Information Processing Series code

nCLOTHIANIDIN The amount in kg of CLOTHIANIDIN applied

nIMIDACLOPRID The amount in kg of IMIDACLOPRID applied

nTHIAMETHOXAM The amount in kg of THIAMETHOXAM applied

nACETAMIPRID The amount in kg of ACETAMIPRID applied

nTHIACLOPRID The amount in kg of THIACLOPRID applied

nAllNeonic The amount in kg of all Neonics applied = (nCLOTHIANIDIN +

nIMIDACLOPRID + nTHIAMETHOXAM + nACETAMIPRID + nTHIACLOPRID)

Table 1. Variables and Descriptions

7

APPENDIX B: TIME SERIES PLOTS

Figure 2. Annual Total Honey Yield in Southern States

Figure 3. Annual Total Honey Yield in Western States

8

Figure 4. Annual Total Honey Yield in Midwestern States

Figure 5. Annual Total Honey Yield in Northeastern State